Pith. sign in

REVIEW 4 major objections 6 minor 27 references

Causally Consistent Normalizing Flow

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read First deep normalizing flow that matches any causal graph without losing expressiveness

desk verdict A genuinely useful construction for deep causally consistent flows, but the paper's headline universality claim rests on a circular proof and needs a proper formalization. read the letter →

arxiv 2412.12401 v1 pith:MKOWNLW4 submitted 2024-12-16 cs.LG

Qingyang Zhou , Kangjie Lu , Meng Xu This is my paper · ORCID
classification cs.LG
keywords causallyconsistentgenerativemodelnormalizingflowstructuralcausalinferencecounterfactualuniversalitytopologicalbatchingfairness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Generative models like normalizing flows can reproduce observed distributions while inducing causal connections that differ from the true structural causal model, a mismatch that can produce unfair decision systems. Earlier fixes achieve causal consistency only by removing hidden layers, which caps the complexity of distributions they can represent. This paper introduces CCNF, a normalizing flow that is causally consistent by construction and, the authors claim, remains a multi-layer universal approximator for essentially any target distribution. The construction reorders the causal graph into a sequence of batches via topological batching and composes partial causal transformations that update one batch at a time using only parent variables as conditions.

What carries the argument

The central construct is the sequential representation of an SCM: a topological batching of the causal DAG into an ordered sequence of variable batches. Carrying the argument is the partial causal transformation T^L_theta, which maps U_i to X_i = T_theta(U_i | X_pa_i) for variables in the label set L and leaves other variables untouched; composing these transformations along the batch sequence yields a triangular flow that is invertible and preserves causality. The proof that this flow is a multi-layer universal approximator is what the paper's claim rests on: it chains the per-batch transformations and invokes the universality of each piece.

What would settle it

Train CCNF on a distribution where a batch's conditional given parents cannot be expressed by the chosen flow layer without sharing latent factors among batch members; if the consistency loss (Equation 7) cannot be driven to zero while approximating the distribution, the per-batch universality assumption fails.

Watch

Extended reading notes

Core claim

The paper claims that CCNF is the first causally consistent generative model that can approximate any distribution with multiple layers. Concretely, for a given structural causal model with causal DAG G, topological batching partitions the variables into an ordered sequence B1,...,Bn. CCNF is the normalizing flow whose transformation is the composition of partial causal transformations $T^{{B_n}}$ ∘ ... ∘ $T^{{B_1}}$, where each $T^{{B_j}}$ updates the variables in batch B_j by the rule X_i = T_theta_j(U_i | X_pa_i) and leaves all other variables unchanged. The paper proves (Theorem 5.1) that this construction makes each variable X_i depend only on its parents and its own exogenous U_i, hence (Theorem 5.3) the induced causal graph matches the SCM; (Theorem 5.2) that the full flow is a multi-layer universal approximator provided each partial transformation is one; and (Theorem 5.4) that it must have at least as many layers as the longest path in the DAG. The paper also gives algorithms that perform observations, interventions, and counterfactuals through this flow, and reports experiments on synthetic benchmarks and the German credit dataset where CCNF maintains causal consistency and improves causal inference accuracy over single-layer baselines.

Load-bearing premise

The per-batch transformation being a multi-layer universal approximator while preserving independence among variables in the same batch is assumed rather than proved for the specific flow layers used in the paper.

Editorial extensions

If this is right

  • Causal consistency and depth are compatible: deep causally consistent models become feasible where previous designs had to flatten to a single layer.
  • One trained CCNF answers all three tiers of Pearl's hierarchy by pushing samples, fixing exogenous variables, and inverting the flow.
  • Because partial transformations act on disjoint batches, the Jacobian determinant stays block-triangular, so exact likelihood training remains tractable as the graph grows.
  • The German credit experiment suggests that a causally consistent flow can zero out individual unfairness while improving classifier accuracy.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's universality proof is conditional: it assumes each partial causal transformation is a multi-layer universal approximator; whether standard coupling layers satisfy this per-batch property is not established, so a concrete verifier (e.g., checking the consistency loss on a crafted conditional) would settle it.
  • If the per-batch property does hold for affine coupling flows, recent coupling-based universality results would transfer directly to CCNF, making the construction a recipe for causally constrained deep generative models beyond flows.
  • The minimum-layer theorem makes the expressiveness/consistency tradeoff explicit: a causally consistent deep model needs at least as many layers as the longest path in the DAG, so the gain over single-layer models is the ability to put many hidden layers inside each per-batch transformation.
  • A natural extension the authors leave implicit is using the consistency loss (Equation 7) as a scoring function for causal discovery, since a model that cannot drive that loss to zero under a candidate graph is evidence against the graph.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces CCNF, a normalizing flow that enforces causal consistency with a given structural causal model by composing 'partial causal transformations' over topologically sorted batches. Each batch transformation maps the exogenous variables of that batch to the corresponding endogenous variables while conditioning on the already-transformed parent variables, and multiple normalizing-flow layers are allowed within each batch. The paper claims that CCNF is the first causally consistent generative model that is a multi-layer universal approximator (Theorem 5.2), and it presents algorithms for observational, interventional, and counterfactual queries. Experiments on synthetic datasets report lower KL/MMD/RMSD errors than CausalNF and VACA, and a German credit application reports elimination of individual unfairness while improving accuracy.

Significance. The sequential-batching construction is elegant, and the causal-consistency-by-design property is a genuine contribution: it shows a simple way to build deep normalizing flows whose induced graph matches a given DAG, which prior causally consistent flows (e.g., CausalNF) achieved only with zero hidden layers. If Theorem 5.2 were rigorously established, CCNF would be a substantial step beyond the state of the art. However, the manuscript does not prove the universality claim: the appendix proof is circular, and the theorem's hypothesis is not formalized for the actual MAF/flow++ architectures used. The empirical causal-consistency score is zero by construction, and the fairness result is engineered through the chosen SCM and manual noise setting. The core theoretical novelty therefore remains unverified, although the construction itself appears sound and likely repairable with a proper universal-approximation theorem for conditional flows.

major comments (4)
  1. [Theorem 5.2 and Appendix B.6] The proof of Theorem 5.2 is circular. It assumes each block T^{B_j} is a multi-layer universal approximator and then concludes 'The T^B_theta is a multi-layer universal approximator since T^B_theta can approximate any P(X)', which is exactly the statement to be proven. More substantively, the theorem's hypothesis is ill-posed: a partial causal transformation acts as X_i = T(U_i | U_pa_i) for i in B_j, so variables within a batch are conditionally independent given their parents. The universal-approximation target must therefore be the family of product conditional distributions prod_{i in B_j} P(X_i | X_pa_i), and the paper does not state or prove that MAF/flow++ with causal masking can approximate arbitrary elements of this product family. Because the abstract's central claim ('first causally consistent generative model that can approximate any distribution with multiple layers') rests on Theorem 5.2, this gap is load-bearing and must be fixed with a precise statement and proof, not a restatement of the assumption.
  2. [Appendix B.7 (proof of Theorem 5.3)] The proof of causal consistency misuses Theorem B.2. It writes X_i = T^{B_j}(U_i), omitting the conditioning on X_pa_i that appears in Theorem 5.1, and then invokes Theorem B.2, which requires X_i to depend only on U_i. Since Theorem 5.1 itself states X_i = T^{B_j}(U_i | X_pa_i), the variable X_i depends on its parents, so the proof as written is invalid. The conclusion may be true for the block-triangular construction, but it needs a direct argument, for example showing that the Jacobian of the full transformation is block-lower-triangular with zero entries for all non-parent edges. As it stands, a central stated guarantee is supported by an incorrect proof.
  3. [Section 6.1 and Table 2] The causal-consistency metric L(T_theta(X)) is reported as 0.00 for CCNF. This is guaranteed by construction: each variable is transformed only in its own batch conditioned on its parents, so the Jacobian has zero entries for all non-parent edges by design. The experiment therefore does not empirically validate causal consistency; it only checks that the implementation matches the design. The paper should acknowledge this and instead validate that the learned conditional distributions are accurate (as attempted in Table 3), or use a metric that can detect implementation errors rather than a quantity that is identically zero for the proposed architecture.
  4. [Section 6.3 and Table 5] The fairness claim is not a learned property of CCNF. The paper states that the classifier sets the exogenous variable of the risk attribute to its mean value (0), and the selected SCM apparently excludes sex as a parent of risk; consequently ATE_gender = 0 and the fairness metric of 0 follow by construction. To substantiate the claim that CCNF 'addresses challenges like unfairness effectively,' the paper needs to show that the learned CCNF itself does not propagate gender to risk through indirect paths, or compare against a baseline that uses the same SCM but without the manual mean-noise intervention. As presented, the result is an artifact of the experimental protocol rather than a demonstration of the model's capability.
minor comments (6)
  1. [Abstract] The phrase 'Struct Causal Models' should be 'Structural Causal Models'.
  2. [Section 3] The term 'casual inconsistency problem' appears in the introduction and elsewhere; it should be 'causal inconsistency problem'.
  3. [Appendix A, Algorithm 2] The Do procedure computes U and then assigns U_i <- T^{B_j,-1}(a | X_pa_i), but the return value S is never defined; presumably it should return the modified X. Please clarify the notation and the output of the procedure.
  4. [General] The paper repeatedly refers to 'the extended version' for proofs, algorithms, and experimental details, but the arXiv submission includes appendices; if this is for a venue with page limits, the extended version should be linked or the missing details should be included so that reviewers can verify the claims.
  5. [Figure 12 caption] The caption says 'These outcomes corroborate the theoretical expectations Figure 5', but Figure 5 does not exist in the paper; the reference should be to the relevant theorem or section.
  6. [Section 4, Running Example] The description of the red arrow says 'the left variable is the dependent variable on the right variable,' which is confusing; the direction of dependence should be described more clearly.

Circularity Check

3 steps flagged · score 7.0 of 10

The central universality proof assumes its own conclusion; the causal-consistency and fairness 'validations' are consequences of the construction, not independent empirical results.

  1. other [Section 5 Theorem 5.2; Appendix B.6 Theorem B.6 proof]
    "Since we assume T^{B_j}_{theta_j} is a multi-layer universal approximator, the distribution P(X_Bj) can be approximated ... The T^B_theta is a multi-layer universal approximator since T^B_theta can approximate any P(X)."

    The proof's final sentence asserts exactly the proposition being proved. The theorem requires showing that composing universal blocks yields a flow universal for the joint P(X); instead the proof assumes each block can approximate its marginal P(X_Bj) and then concludes the composition can approximate any P(X) without deriving the conditional/product structure. The claimed universality of the concrete MAF/flow++ blocks is also never established, so the abstract's 'first ... approximate any distribution with multiple layers' rests on an assumed antecedent.

  2. self definitional [Section 6, Causal Consistency, Result]
    "Firstly, CCNF demonstrates causally consistent with the given SCM, as L(T_theta(X)) of CCNF is consistently 0."

    Equation 7 defines L(T_theta(X)) = ||nabla_x T_theta(X) . (1-G)||_2. For CCNF the Jacobian is block-lower-triangular with zero blocks exactly where G has no edges, by the definition of partial causal transformations and topological batching. Hence L=0 holds for any parameters; reporting it as an experimental result verifies the construction against itself, not a learned property.

1 more flagged steps
  1. self definitional [Section 6, Real-world Evaluation; Appendix D, ITE and ATE Evaluation]
    "CCNF could function as an unfairness-free classifier by setting the exogenous variable of the risk attribute to its mean value, which is 0 in our experiment. ... since the age and gender are not directly used when computing the risk, changing the value directly cannot affect the value of risk."

    Individual fairness is defined as (1/n) sum |Risk(sex=1)-Risk(sex=0)|. The German credit SCM used by CCNF has no parent edge from sex to risk, and the risk exogenous variable is set to its mean (0). Therefore the do(X_sex=1) and do(X_sex=0) distributions of risk are identical by construction, so the reported 0.00 unfairness is an identity of the chosen SCM and noise assignment, not an empirical demonstration that CCNF removes bias.

full rationale

CCNF's causal consistency is genuinely by construction: blockwise partial causal transformations with topological batching force the Jacobian to respect the graph. That part is not circular as a design guarantee. But the paper presents this algebraic identity as an empirical result (L=0), which is a tautology. The load-bearing universality claim is not established: Theorem B.6's proof assumes each block is a multi-layer universal approximator and then concludes the composition is universal because 'T^B can approximate any P(X)'—the very statement to be proved. The concrete MAF/flow++ blocks are conditional transformations whose conditional-universality property is never proved; citing unconditioned MAF universality does not cover it. The fairness case study likewise fixes risk's exogenous noise to its mean in an SCM where sex is not a parent of risk, so the zero unfairness is forced by the setup, not a measured effect of CCNF. No load-bearing self-citations are involved; the circularity is definitional and assumptional.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard SCM assumptions (DAG, independent exogenous noise), a known topological sorting lemma, and an unproven universality assumption for each partial causal transformation block. The fairness result additionally depends on a hand-chosen causal graph and a hand-set exogenous value. No new physical or ontological entities are introduced.

free parameters (2)
  • Risk exogenous variable set value in fairness classifier = 0
    In the German credit experiment, CCNF is used as a fairness classifier by setting the exogenous variable of the risk attribute to its mean, 0. This hand-chosen value forces the reported 0.00 individual unfairness.
  • Layer count per partial causal transformation block = 1 to 5
    The length of each T^B_theta varies from 1 to 5, tuned with Optuna. This is a hyperparameter that affects expressiveness and the multi-layer claim.
assumptions (5)
  • domain assumption Causal graph G is a DAG and is known in advance.
    Assumed in Section 3 and used by topological batching; the authors acknowledge in Future Work that requiring a well-defined causal graph is a major constraint.
  • domain assumption Exogenous variables U_i are mutually independent.
    Standard SCM assumption stated in Section 3; it underlies the factorization of batch conditionals and the proof of causal consistency.
  • standard math Lemma B.1: parents of a node in batch B_j are contained in earlier batches.
    Standard property of topological sorting, cited from Thost and Chen 2021; used in Theorem 5.1 to identify Upa_i with Xpa_i.
  • ad hoc to paper Each partial causal transformation T^{B_j} is a multi-layer universal approximator for the conditional distribution of its batch given its parents.
    Theorem 5.2 assumes this premise and Appendix B.6 does not prove it; the universality claim of CCNF depends on this unverified premise.
  • domain assumption Theorem B.2 from Javaloy et al. correctly characterizes causal consistency via isolation of exogenous variables.
    The proof of Theorem B.7 invokes this prior theorem but restates it as 'X_i only depends on U_i', which would erase the intended causal edges; the valid statement of the theorem is essential.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causally Consistent Normalizing Flow." pith.science (2026). https://pith.science/paper/MKOWNLW4

@misc{pith2026241212401,
  author       = {Pith},
  title        = {Pith review of: Causally Consistent Normalizing Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MKOWNLW4}},
  note         = {Machine review of arXiv:2412.12401}
}
read the original abstract

Causal inconsistency arises when the underlying causal graphs captured by generative models like \textit{Normalizing Flows} (NFs) are inconsistent with those specified in causal models like \textit{Struct Causal Models} (SCMs). This inconsistency can cause unwanted issues including the unfairness problem. Prior works to achieve causal consistency inevitably compromise the expressiveness of their models by disallowing hidden layers. In this work, we introduce a new approach: \textbf{C}ausally \textbf{C}onsistent \textbf{N}ormalizing \textbf{F}low (CCNF). To the best of our knowledge, CCNF is the first causally consistent generative model that can approximate any distribution with multiple layers. CCNF relies on two novel constructs: a sequential representation of SCMs and partial causal transformations. These constructs allow CCNF to inherently maintain causal consistency without sacrificing expressiveness. CCNF can handle all forms of causal inference tasks, including interventions and counterfactuals. Through experiments, we show that CCNF outperforms current approaches in causal inference. We also empirically validate the practical utility of CCNF by applying it to real-world datasets and show how CCNF addresses challenges like unfairness effectively.

Figures

Figures reproduced from arXiv: 2412.12401 by the authors.

Figure 1
Figure 1. Prior distributions of causal NF and CCNF and formally prove its ability to maintain the causal con￾sistency. • Utilizing this sequential representation alongside partial causal transformations, we develop CCNF, a GM that guar￾antees causal consistency and excels at complex causal inference tasks. • We empirically validate CCNF, demonstrating it outper￾forms state-of-the-art casually consistent GMs on causal inferen… view at source ↗
Figure 2
Figure 2. The causal graph of an SCM describing an admis [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Causally consistent models and inconsistent mod [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: An example SCM, its topological order, and a related [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: Causal graph for Four Node Chain dataset [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Causal graph for Four M-Graph dataset 𝑋0 𝑋1 𝑋2 𝑋3 𝑋4 Network. The Network dataset consists of 6 endogenous variables. It simulates a 2-layer neuron network with softplus activate layer. The structural equations and the causal graph are list below. 𝑋0 = 𝑈0 𝑋1 = 𝑆𝑜 𝑓 𝑡 𝑝…
Figure 10
Figure 10. Figure 10: Causal graph for Backdoor dataset 𝑋1 𝑋2 𝑋4 𝑋0 𝑋6 𝑋3 𝑋5 Eight Node Chain. The Eight Node Chain dataset consists of 8 endogenous variables. This dataset is used in the causal inference tasks experiment. The structural equations and the causal graph are list below. 𝑋0 = …
Figure 12
Figure 12. Figure 12: Intervened distributions of 𝑁𝑙𝑖𝑛 − 𝑇𝑟𝑖𝑎𝑛𝑔𝑙𝑒 on CCNF, VACA, causalNF. 8], [16, 16], [32, 32], [64, 64]}. However, unlike others, the length of each 𝑇 B 𝜃 in CCNF varies from 1 to 5. Training methodology. The synthetic dataset is generated through the following process:…
Figure 5
Figure 5. Figure 5: Since CCNF is a multi-layer universal approxima￾tor, it possesses the capability to capture the true generative process regardless of its complexity, hence performing well on both intervened distributions. CausalNF limits its depth for maintaining causal consistency, r…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 8 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Agarap, A. F. 2018. Deep learning using rectified linear units (relu). arXiv preprint arXiv:1803.08375

    arXiv 2018

  4. [4]

    Akiba, T.; Sano, S.; Yanase, T.; Ohta, T.; and Koyama, M. 2019. Optuna: A next-generation hyperparameter optimization framework. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, 2623--2631

    2019

  5. [5]

    Crouse, M.; Abdelaziz, I.; Cornelio, C.; Thost, V.; Wu, L.; Forbus, K.; and Fokoue, A. 2019. Improving graph neural network representations of logical formulae with subgraph pooling. arXiv preprint arXiv:1911.06904

    work page Pith review arXiv 2019

  6. [6]

    o rr, C.; and K \

    Draxler, F.; Wahl, S.; Schn \"o rr, C.; and K \"o the, U. 2024. On the universality of coupling-based normalizing flows. arXiv preprint arXiv:2402.06578

    arXiv 2024

  7. [7]

    Fleisher, W. 2021. What's fair about individual fairness? In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society, 480--490

    work page 2021

  8. [8]

    K.; Pandey, M.; and Rautaray, S

    Harshvardhan, G.; Gourisaria, M. K.; Pandey, M.; and Rautaray, S. S. 2020. A comprehensive survey and analysis of generative models in machine learning. Computer Science Review, 38: 100285

    work page 2020

Show all 27 references
  1. [9]

    Hofmann, H. 1994. Statlog (German Credit Data) . UCI Machine Learning Repository. DOI : https://doi.org/10.24432/C5NC77

    1994 doi

  2. [10]

    S.; and Valera, I

    Javaloy, A.; Martin, P. S.; and Valera, I. 2023. Causal normalizing flows: from theory to practice. In Thirty-seventh Conference on Neural Information Processing Systems

    2023

  3. [11]

    P.; Leech, R.; and Hyvärinen, A

    Khemakhem, I.; Monti, R. P.; Leech, R.; and Hyvärinen, A. 2021. Causal Autoregressive Flows. arXiv:2011.02268

    2021 arXiv

  4. [12]

    P.; and Ba, J

    Kingma, D. P.; and Ba, J. 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980

    2014 arXiv

  5. [13]

    G.; and Vishwanath, S

    Kocaoglu, M.; Snyder, C.; Dimakis, A. G.; and Vishwanath, S. 2017. CausalGAN: Learning Causal Implicit Generative Models with Adversarial Training. arXiv:1709.02023

    2017 arXiv

  6. [14]

    Komanduri, A.; Wu, X.; Wu, Y.; and Chen, F. 2024. From Identifiable Causal Representations to Controllable Counterfactual Generation: A Survey on Causal Generative Modeling. arXiv:2310.11011

    2024 arXiv

  7. [15]

    J.; Mohamed, S.; and Lakshminarayanan, B

    Papamakarios, G.; Nalisnick, E.; Rezende, D. J.; Mohamed, S.; and Lakshminarayanan, B. 2021. Normalizing flows for probabilistic modeling and inference. Journal of Machine Learning Research, 22(57): 1--64

    2021

  8. [16]

    Papamakarios, G.; Pavlakou, T.; and Murray, I. 2018. Masked Autoregressive Flow for Density Estimation. arXiv:1705.07057

    2018 arXiv

  9. [17]

    Pawlowski, N.; Coelho de Castro, D.; and Glocker, B. 2020. Deep structural causal models for tractable counterfactual inference. Advances in neural information processing systems, 33: 857--869

    2020

  10. [18]

    Pearl, J. 2009. Causality. Cambridge university press

    2009

  11. [19]

    Pearl, J. 2012. The do-calculus revisited. arXiv preprint arXiv:1210.4852

    2012 arXiv

  12. [20]

    Ribeiro, F. D. S.; Xia, T.; Monteiro, M.; Pawlowski, N.; and Glocker, B. 2023. High Fidelity Image Counterfactuals with Probabilistic Causal Models. arXiv:2306.15764

    2023 arXiv

  13. [21]

    Sanchez, P.; and Tsaftaris, S. A. 2022. Diffusion causal models for counterfactual estimation. arXiv preprint arXiv:2202.10166

    2022 arXiv

  14. [22]

    S \'a nchez-Martin, P.; Rateike, M.; and Valera, I. 2022. VACA: Designing variational graph autoencoders for causal queries. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 8159--8168

    2022

  15. [23]

    Thompso, E. 2023. Class-action lawsuit against Facebook claiming discrimination gets the green light. CBC

    2023

  16. [24]

    Thost, V.; and Chen, J. 2021. Directed Acyclic Graph Neural Networks. arXiv:2101.07965

    2021 arXiv

  17. [25]

    Xi, Q.; and Bloem-Reddy, B. 2023. Indeterminacy in generative models: Characterization and strong identifiability. In International Conference on Artificial Intelligence and Statistics, 6912--6939. PMLR

    2023

  18. [26]

    Xia, K.; Pan, Y.; and Bareinboim, E. 2022. Neural causal models for counterfactual identification and estimation. arXiv preprint arXiv:2210.00035

    2022 arXiv

  19. [27]

    Yang, M.; Liu, F.; Chen, Z.; Shen, X.; Hao, J.; and Wang, J. 2021. Causalvae: Disentangled representation learning via neural structural causal models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 9593--9602

    2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.